Abstract

We present SIRFS (shape, illumination, and reflectance from shading), the first unified model for recovering shape, chromatic illumination, and reflectance from a single image. Our model is an ex- tension of our previous work [1], which addressed the achromatic ver- sion of this problem. Dealing with color requires a modified problem formulation, novel priors on reflectance and illumination, and a new op- timization scheme for dealing with the resulting inference problem. Our approach outperforms all previously published algorithms for intrinsic image decomposition and shape-from-shading on the MIT intrinsic im- ages dataset [1, 2] and on our own “naturally” illuminated version of that dataset.